InfNeRF: Towards Infinite Scale NeRF Rendering with O(log n) Space Complexity (Siggraph Asia 2024)

| Project Page | Paper |

Jiabin Liang¹, Lanqing Zhang¹, Zhuoran Zhao^1,2, Xiangyu Xu³

¹ Sea AI Lab ² National University of Singapore, ³ Xi'an Jiaotong University

Overview

• InfNeRF extends the proven LoD technique to Neural Radiance Fields (NeRF) by introducing an octree structure to represent the scenes in different scales.
• This innovative approach provides a mathematically simple and elegant representation with a rendering space complexity of $\mathcal{O}(\log n)$, aligned with the efficiency of mesh-based LoD techniques.
• We also present a novel training strategy that maintains a complexity of $\mathcal{O}(n)$. This strategy allows for parallel training with minimal overhead, ensuring the scalability and efficiency of our proposed method.
• Our contribution is not only in extending the capabilities of existing techniques but also in establishing a foundation for scalable and efficient large-scale scene representation using NeRF and octree structures.

Demo

Result of Window of the World, ShenZhen, rendering with < 17% of the model:

Result of UrbanScene3D Residence, rendering with < 16% of the model:

Result of UrbanScene3D Sci Art:

Result of Mill 19 Building:

Result of Mill 19 Rubble:

Refer to our project page for more high-resolution rendering results.

Data Preparation

Mill 19

• The Building scene can be downloaded here.
• The Rubble scene can be downloaded here.

UrbanScene 3D

Download the raw photo collections from the UrbanScene3D dataset

• Residence
• Sci-Art
• Campus

After downloading all the raw images, use COLMAP to obtain the camera poses:

ns-process-data images --data ./data/building-pixsfm/data/images --output-dir ./data/building-pixsfm/data --sfm-tool colmap --skip-image-processing --gpu

We have provided the COLMAP results for the Residence dataset: Google Drive. The data structure for InfNeRF training would be like:

- Residence
  - sparse
    - 0
      - cameras.bin
      - images.bin
      - points3D.bin
      - project.ini
  - images
    - A
      - DJI_0413.JPG
      ...
    - B
      - DJI_0001.JPG
      ...
    - C
      - DJI_0001.JPG
      ...

Install

Refer to the nerfstudio environment installation: Installation.

Training

Registering infnerf dataparser with nerfstudio:

pip install -e .

Training command:

ns-train inf-nerf --data ./Residence

You can use tensorboard to see the visualization of the evaluation results and metrics:

tensorboard --logdir=./outputs/Residence/inf-nerf/2025-01-06_143012 --port=6010

Citation

If you find this project useful, please consider citing:

@inproceedings{10.1145/3680528.3687646,
    author = {Liang, Jiabin and Zhang, Lanqing and Zhao, Zhuoran and Xu, Xiangyu},
    title = {InfNeRF: Towards Infinite Scale NeRF Rendering with O(log n) Space Complexity},
    year = {2024},
    url = {https://doi.org/10.1145/3680528.3687646},
    doi = {10.1145/3680528.3687646},
    booktitle = {SIGGRAPH Asia 2024 Conference Papers},
}